The Week the Apocalypse Didn't Show Up: Four Major Outages in Six Days, Zero Attacks. Every One Was Ordinary Maintenance Going Wrong, and That Is Where Your Budget Is Missing

Spend one hour this week on your own change process before you spend another dollar on the next big threat. Between Thursday, November 12 and Tuesday, November 17, 2026, email, air travel, banking and the NSA's classified network all went down. Not one of those outages is reported as an attack. Every one, as far as anyone has said, was routine work on the system going wrong.
The scoreboard
- Thursday: Microsoft 365. A latent bug from a July fix, set off by a traffic surge on a fleet with reduced capacity. Microsoft: "This was not a cyberattack" (via Fox Business). Our post.
- Friday: the FAA. Data flooding its telecom network from two remote sites, one in the middle of a router replacement. About 12,462 flights canceled. The FAA: "no indication that this was a cyberattack." Our post.
- Monday: Chase, and the banks connected to it through Zelle. A software problem, per Chase via the press. Our post.
- Tuesday: NSA Net, the agency's classified internal network. Routine patching or software upgrades, per anonymous sources. No evidence of a cyberattack, per officials who spoke to CBS. Our post.
Attacks: zero. Self-inflicted: four.
Meanwhile, in the same week
On November 16, Google's threat intelligence team published research titled "Agentic threat activity crosses a significant threshold," and BleepingComputer's write-up of Google's warning ran under the headline "Google warns that fully-automated AI attacks are coming soon." Google Project Zero cut its disclosure deadline to 30 days, citing AI. We covered the Google research the same day, and it is solid: an attacker framework turning patches into working exploits, and a North Korea-linked group planting backdoors through Claude Code settings files. Our post on that.
None of that is wrong. AI is making attackers faster, and patch windows have to shrink. But look at what actually took the country offline this week: a gateway fix from July, a router swap at 3 p.m. on a Friday, a bank software change and a patch cycle. The apocalypse everyone was briefed on did not show up. The change window did, four times.
Why the budget misses this
Security spending is shaped by fear of an adversary, and most of the product market sells against one: detection, response, threat feeds (including ours). That money buys defense against the confidentiality and integrity failures attackers cause.
Availability fails from the inside far more often, and the defenses are not products. They are habits:
- Rollback you have actually tested, including for changes that have been live for months. Microsoft's July bug could not be rolled back quickly because by November it was the baseline.
- Headroom, counted with today's maintenance subtracted. Take 14 percent of a fleet offline and every remaining server carries 16 percent more load before any surge arrives.
- Walls, so one site's problem stays at one site. One FAA site's traffic reached 17 of 22 en route centers.
- A slow, staged recovery. Microsoft's servers fell over again as they came back. The NSA reportedly staggered staff arrivals so everyone would not log in at once.
- Memory. The FAA had nearly the same failure in 2009. A router configuration copied from the replaced router, errors included, took down 129 facilities. The Inspector General warned Congress in 2010 that the FAA was "reactive, rather than proactive." Sixteen years later it happened again. The full 2009 story.
None of these costs real money. They cost attention, which is harder to buy and impossible to outsource.
Blame the STIG, not the soldier
The people who did these changes were, as far as anyone knows, doing their jobs: applying a fix, swapping a router, running a patch cycle. The rules they work under are mostly about the destination. Hardening guides like the Defense Department's STIGs, and patch deadlines like CISA's known-exploited list, say what state a system must reach and by when. Far fewer rules say how to get there without taking the system down: test the rollback, check the headroom, keep the change behind a wall, bring people back slowly.
So the operator gets a deadline and a checklist, and the safety of the change is left to whoever is on shift. When it breaks, the after-action report names a technician. In 2009 it named a Harris engineer who missed one line of a router configuration. The process that let one missed line reach 129 facilities stayed, and sixteen years later the boot filled up again. Fix the process and the person stops being the single point of failure.
This is not an argument against threat intelligence
We sell threat intelligence, so take this for what it is: the honest version. Attackers are real, AI is making them faster, and you should patch faster than you did last year. Our feed exists because of that.
But a security program that covers adversaries and ignores its own change process has covered half the risk. This week showed which half causes the outages. If your incident playbook starts with "is it an attack?", add a second line: "what did we change?" On the evidence of this week, the second question finds the answer first.
What to do this week
- Pull the last five changes to anything customers or staff depend on. For each one, write down whether a rollback was tested and who was watching while it happened.
- Count your real spare capacity during your next maintenance window, not on a normal day.
- Decide how users reconnect after an outage: in waves, with clients that back off.
- Ask your vendors, cloud, bank and payment rails included, for the post-incident review when they fail. If they will not give you one, that tells you something too.
What is not known
Every root cause above is preliminary. Microsoft has promised a full review; the FAA and L3Harris are still investigating; Chase's account comes through the press; the NSA's cause comes from anonymous sources. If any of them turns out to be an attack after all, we will say so here, in this post.
Sources
Each linked post carries its full citations. The headlines quoted here: Google Threat Intelligence Group, "Agentic threat activity crosses a significant threshold," November 16, 2026; BleepingComputer, "Google warns that fully-automated AI attacks are coming soon," November 16, 2026; BleepingComputer, "Google Project Zero cuts bug disclosures to 30 days amid AI threats," November 16, 2026. DOT Office of Inspector General, letter CC-2010-012, June 17, 2010.
How do AI models see YOUR brand?
AIPM has audited 250+ domains. 15 seconds. Free while still in beta.
Audit your site now → aipmsec.com
Was this useful? Thirty seconds, no cookies, no tracking, no third parties, your address hashed and never stored. The question lives here: https://analytics.dugganusa.com/nps.html#post=week-the-apocalypse-didnt-show-up